An intelligent identification model for substation flood risk based on LightGBM and AHP weight optimization

Extreme rainfall events driven by global climate change pose severe challenges to the safety and stability of power transmission systems. Traditional analytic hierarchy process (AHP)-based methods for substation flood-risk assessment suffer from subjective weight assignment and poor adaptability to nonlinear data relationships. To address these issues, this study develops an intelligent flood-risk identification model integrating the AHP and the Light Gradient Boosting Machine (LightGBM) through dynamic weight optimization. AHP establishes a hierarchical indicator framework and provides initial subjective weights, while LightGBM learns objective feature importance from multi-source data collected from 45 substations in Henan Province. The final weights are obtained through linear fusion of expert-based and data-driven results. Experimental results show that the proposed AHP–LightGBM model achieves an average accuracy of 91.2%, improving by 15% over the traditional AHP model and by 6.8% over the standalone LightGBM model. Rainfall intensity, drainage capacity, terrain elevation, and equipment sealing depth are identified as dominant risk factors. The model demonstrates strong interpretability, adaptability, and generalization capability, providing a reliable framework for intelligent flood-risk assessment and decision-making in power grid systems.